Economic Data Revisions: Why the Previous Number Changes

A jobs trend can weaken without a single new bad month being reported. That happens through revisions. Agencies publish their first estimates quickly, while many survey responses are still missing, and then replace those figures as late responses and fuller data come in. So the Previous column on an economic calendar often shows a revised number, different from the one you saw a month ago. Some calendars still show the original. Either way, the change is information.

Why are first estimates incomplete?

Speed has a cost. A monthly jobs report comes out within weeks of the month’s end, and plenty of the businesses in the survey haven’t answered by then, so the agency estimates from the responses it has and folds in the late ones when they arrive.

GDP works the same way. It draws on a wide range of source data, some of which only exists months after the quarter closes, so the early estimates fill those gaps with projections and the later ones swap in the real figures as each source reports. The gaps shrink each time.

The first number was the best one available at the time, published fast because a rough answer early is worth more to most users than a precise one late. Revisions are how the agencies catch up with the data that arrived after the deadline, on a timetable they set out ahead of time.

How do the agencies schedule revisions?

The schedule is published. You can know in advance which numbers are still provisional.

So any payrolls month gets revised twice in the reports that follow it. The benchmark can move it again. Most calendars show only one previous figure on the payrolls row, the month just before, so the revision to the month before that is easy to miss unless you open the release itself.

GDP shows up three times per quarter. The advance estimate gets the attention. The second and third are the same quarter, measured again with more of the data in hand.

What does a revision do to the trend?

It reshapes the recent past. Take three hypothetical months of payrolls.

No new month came in weak. The latest print hasn’t moved. But a reader who judged the trend from the first estimates saw steady hiring near 160,000 a month, and after two rounds of revision the same three months average under 147,000, a softer picture that only shows up if you go back and recompute.

Read the revision with the headline. The actual vs forecast walkthrough nets a revision against a headline miss on a single release day.

Why does a revision move prices on release day?

Because the market had already priced the old number. Last month’s first estimate went into expectations for interest rates, for consumer spending and for company sales, and when the agency changes it, those expectations have to change with it in the same minute the new month prints. A headline that matches its forecast can still move stocks if the month before it was cut. The reverse holds too.

The effect is larger when the revision runs the same way as the surprise. A miss on the new month plus a downward revision to the old one says the same thing twice. A miss offset by an upward revision says something milder.

Which previous figure should you compare with?

The revised one. Measure the new month against the current estimate of the month before. Comparing with the original first estimate mixes an old view of last month with a current view of this one, and that comparison can make a flat month look like a slowdown or a pickup.

Using the revised figures above, a new month of 150,000 would match the revised second month exactly. Against that month’s first estimate of 170,000 it would look like a 20,000 drop. Same print. Only one comparison uses current data.

What does the direction of revisions tell you?

One revision tells you little. Samples bounce. A run of revisions in one direction says more.

If the first estimates have been revised down release after release, the early numbers have been running too high, and the next first estimate may be flattering the economy as well. A string of upward revisions points the other way. Either pattern is a reason to hold the latest headline more loosely, though neither predicts the next print.

Tracking it is cheap. Keep a simple table with the release, the first estimate and each revision as it arrives, and after a few months the direction of the changes is visible at a glance without any further arithmetic. A column for the latest figure and one for the first print do most of the work.

Weekly series carry a previous figure too. For initial jobless claims, check the release for whether the prior week’s number has moved. That matters most when hiring is the worry.

Does any of this matter for a stock position?

It matters through rates and earnings. A trend that turns out weaker than first reported can change what investors expect a central bank to do, and it can change what analysts expect for companies that sell to consumers or employ many people, both of which feed into the prices of stocks you may hold. For a swing trade lasting days, one revision rarely matters. For a position held through several quarters, the revised trend is the one that counts.

How does this fit the rest of the calendar?

Revisions are one column of a row that also holds the actual and the forecast, and reading all three together is the method set out in how to read an economic calendar. For how jobs, inflation and output feed through to company earnings and stock valuations, the free courses include a full course on macro data, with a lesson on the economic data that reaches earnings.